Subhashish Satpathy

Key Roles and Responsibilities

• Provided technical guidance and support to development teams in the implementation of architectural designs.
• Performed code reviews and debugging of applications to ensure quality standards.
• Collaborated with other architects and developers to evaluate system requirements, analyze solutions, and develop strategies.
• Developed and implemented technical architecture blueprints for various projects.

ICD and HCC Code Extraction.

• Developed a multi-class, multi-label classification model using BERT and BigBird architectures. Optimized model performance by running over 250 experiments. Implemented a scalable inference pipeline using GCP DocAI Processor for real-time predictions.

Pharmaco Vigilance ABON Causality Prediction

• Trained an XGBoost model and improved its performance through hyperparameter tuning. Deployed the model using Azure MLOps for operationalization.

Document Ingestion and Knowledge Base Creation.

• Led a team in designing and implementing a high-performance data ingestion pipeline. Successfully transferred and processed 10 million documents. Integrated documents into Vertex AI Agent Builder for easy search and retrieval.

Regulatory Data Scraping and Summarization.

• Developed a robust web scraping pipeline to extract regulatory data. Implemented efficient algorithms for data comparison. Leveraged Gemini's capabilities to generate accurate summaries.

Revenue Forecasting

• Analyzed the time series characteristics of the data, i.e., stationarity, seasonality, and trend. Built an ARIMA model.

Entity Extraction from Legal Documents

• Designed and trained an NER model using BiDirectional LSTM + CRF model (weighted F1 89%), along with rule-based methods.

Clause Category and Sub-Category Classification

• Built a classification model to accurately (weighted F1-score ~92%) categorize clauses based on their content.

Intent Classification of Japanese Healthcare Data

• Created an LSTM-based classification model. Handled the challenges of Japanese text tokenization.

Title Generation for Japanese Article

• Implemented a pointer generator network to generate titles that accurately summarize the article content.

Resume Parsing

• Performed IOEB tagging. Trained a NER model (weighted F1 ~89%).

• Troubleshooting testing, and debugging COBOL and JCL errors.